Not anything useful below 50kHz but certainly improves the lower region.
If there is no error in my calculation
1), then 1mOhm shunt has S21 of -91dB, and 0.1 mOhm -111dB.
Noise floor is still too high for that use case. Ideally, you'd rather need a noise floor of say -120dB, or even lower.
You could try to increase the averaging further to 100x or 1000x. Rather reduce the number of points, if it becomes too slow.
But even if you can further reduce the noise floor, I wonder whether the gain calibration will be accurate enough at such low levels.
The discontinuity at 400 kHz is quite large, too.
1) https://cds.cern.ch/record/1415639/files/p67.pdf, page 88, "3. shunt admittance", for Y1=Y2=1/50, and Y=1kS and Y=10kS